review-anvil-improve-pr

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill workflow reads GitHub PR history, comments, and repository state via gh CLI commands and helper scripts, exposing the agent to outsider-authored free text from GitHub PR comments and code contributions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The skill instructs installation of external skills via npx skills add mrshu/agent-skills, which is a remote-repository locator/package manager invocation fetching code and instructions at runtime, but evaluation yields no verifiable official provenance for the third-party GitHub user/organization repository mrshu/agent-skills, categorizing it as a weak/unknown/suspicious operational dependency.